Since tzdata-2015e there are UTF-8 characters in iso3166.tab, see:
http://mm.icann.org/pipermail/tz/2015-May/022258.html
http://mm.icann.org/pipermail/tz/2015-June/022306.html

pytz/__init__.py:_CountryNameDict(LazyDict) is using:
        zone_tab = open_resource('iso3166.tab')
        try:
            for line in zone_tab.readlines():
                line = line.decode('US-ASCII')

to read it and fails on AX, CI, RE lines, using UTF-8 fixes the
issues and should work OK even with older tzdata releases.

diff -uNr pytz-2015.4.orig/pytz/__init__.py pytz-2015.4/pytz/__init__.py
--- pytz-2015.4.orig/pytz/__init__.py	2015-05-13 03:50:26.000000000 -0700
+++ pytz-2015.4/pytz/__init__.py	2015-07-10 11:35:30.080242726 -0700
@@ -357,7 +357,7 @@
         zone_tab = open_resource('iso3166.tab')
         try:
             for line in zone_tab.readlines():
-                line = line.decode('US-ASCII')
+                line = line.decode('UTF-8')
                 if line.startswith('#'):
                     continue
                 code, name = line.split(None, 1)